Case Study

Decide how a reviewer should document a batch of misclassified dev set images.

Case context: A reviewer plans to examine about 100 misclassified dev set images. The reviewer wants to record recurring categories while also retaining details that make individual examples memorable.

Question: What documentation approach should the reviewer use, and which fields should it include according to the source?

Sample answer: The reviewer should create and fill out a spreadsheet while examining the images. It should include fields for Image, Dog, Great cat, Blurry, Mislabeled, and Comments. The category fields structure the observations, while the Comments field captures details that help the reviewer remember specific examples.

Key points:

  • Create the spreadsheet for the review.
  • Fill it out while examining about 100 misclassified images.
  • Include Image, Dog, Great cat, Blurry, Mislabeled, and Comments fields.
  • Use comments to preserve memorable details about individual examples.

Rubric: The response should recommend a spreadsheet used during the review, identify the source-supported fields, and explain the role of comments. Do not award full credit for an answer that omits either the structured fields or example-specific comments.

0

1

Updated 2026-07-19

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI